Feature Selection for high Dimensional DNA Microarray data using hybrid approaches.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• DNA microarray data presents high dimensionality challenges, with more genes than samples, making direct classifier training difficult.
• Identifying a small subset of genes strongly correlated with specific classes is essential but often non-trivial.
• Robust and reliable gene finding methods are needed for effective analysis of gene expression data.

Purpose of the Study:

• To develop and evaluate hybrid feature selection approaches for identifying relevant genes in DNA microarray data.
• To investigate the efficacy of combining filtering and wrapper methods for gene selection.
• To assess the performance of specific filtering (Information Gain, Pearson Product Moment Correlation) and wrapper (Biogeography-Based Optimization) techniques.

Main Methods:

• Hybrid feature selection combining filtering and wrapper phases.
• Filtering parameters: Information Gain (IG) and Pearson Product Moment Correlation (PPMC).
• Wrapper approach: Biogeography-Based Optimization (BBO).
• Gene subset evaluation using K-Nearest Neighbors (KNN) and Back Propagation Neural Network (BPNN).

Main Results:

• The Information Gain-Biogeography-Based Optimization-K-Nearest Neighbors (IG-BBO-KNN) combination demonstrated impressive performance.
• This hybrid approach achieved high accuracy (over 90%) and low error rates across different datasets.
• The study highlights the effectiveness of hybrid feature selection for gene finding in high-dimensional expression data.

Conclusions:

• Hybrid feature selection methods, particularly IG-BBO-KNN, offer a robust solution for gene finding in DNA microarray data.
• The proposed approach effectively addresses the challenge of high dimensionality and small sample size.
• Accurate gene identification using these methods can significantly improve classifier performance in biological studies.